How to Price Generative AI
I was recently jamming with my good friend (and former?Hustle-staffer) Trung T. Phan about generative AI pricing. Trung launched an app called?Bearly.ai?(it’s dope —?check it out), and helped narrow down a list of companies offering generative AI features in a variety of ways.
Here’s a breakdown of how this list is currently charging:
Three things stick out to me:
领英推荐
I expect this pricing to evolve quite a bit, and sure it already has since hitting publish on this post.
While "words generated" might not be the most intuitive way to charge for this functionality, "per user" pricing has its own challenges.
If Generative AI has the ability to replace human workers (e.g., copywriters, etc.), then charging on a "per user" introduces perverse incentives -- the better the tool works, the less workers will be using it.
With this in mind, it seems clear that differentiation will be key for statups offering Generative AI startups, as much of the value will depend on what vertical the product helps with, and what tasks it helps simplify.
Business Development
1 年A potential solution to the pricing dilemma presented by generative AI tools could be implementing a tiered pricing structure based on output quality or customization level rather than arbitrary metrics like words generated or number of users. This way, customers can choose a plan tailored to their specific needs and derive maximum value from the tool.
CEO Ibbaka Performance - Leader LinkedIn Design Thinking Group - Generative Pricing
1 年I am pretty sure these companies are paying for API access, which is priced differently. It is basically the sum of tokens input and tokens output. Open.ai prices input tokens at 50% of what they price output tokens. Cohere prices input and output tokens at the same level.
I Dispel B2B SaaS Pricing Illusions
1 年I'm not a fan of "words generated" as you end up not using a lot of what is produced, and, as a user, you can't know that in advance. I've also found that GPT-4 tends to be quite verbose, even for simple prompts. If you wanted the "words generated" to work you'd have to have some sort of pre-config set of steps (or cycle) that lets the user confirm the "AI gets it" before generating a long-form answer.
Founder @Brance Global | Building #ServiceNow Teams Globally | Hiring ServiceNow Delivery Director & ServiceNow Practice Head | #ServiceNow Certified- CTA, CMA, ITSM, ITAM, AIOps, ITOM, SecOps, GRC, IRM, HRSD, CSM, FSM
1 年Jasper is well trained and that's it's super expensive
Pricing Research | Customer Insights | Economics
1 年The tricky part is figuring out how to scale pricing to value-in-use, without discouraging use.